PointCaM

This repository is for PointCaM introduced in the following paper:

PointCaM: Cut-and-Mix for Open-Set Point Cloud Learning
Jie Hong*, Shi Qiu*, Weihao Li, Saeed Anwar, Mehrtash Harandi, Nick Barnes, Lars Petersson

Paper and Citation

The paper can be downloaded from arXiv.
If you find our paper/code is useful, please cite:

    @article{hong2023pointcam,
      title={PointCaM: Cut-and-Mix for Open-Set Point Cloud Learning},
      author={Hong, Jie and Qiu, Shi and Li, Weihao and Anwar, Saeed and Harandi, Mehrtash and Barnes, Nick and Petersson, Lars},
      journal={arXiv preprint arXiv:2212.02011},
      year={2023}
    }

Datasets and Environments

PointTransformer:
Download the datasets and set the environments following the project point-transformer. The experiments are running on 4 NVIDIA GeForce RTX 3090.

Running the Code

PointTransformer:

Training
Name exp of exp_dir in ./tool/train.sh;
Configure data_root, test_list, test_list_full, and names_path in ./config/s3dis/s3dis_pointtransformer_repro.yaml;
Configure cutmix, data_split, open_eval, alpha, and select_ratio in ./config/s3dis/s3dis_pointtransformer_repro.yaml.

sh train.sh

Testing
Name exp of exp_dir in ./tool/test.sh;
MSP and MaxLogits methods share the same training process;
Even open_eval is set to "msp" in the training stage,
configure open_eval to "maxlogit" for evaluating MaxLogits.

sh test.sh

Pre-trained models

PointTransformer:
The pre-trained models can be downloaded from here.

ShiQiu0419 / pointcam